— Blog
Notes from the AI-CAD trenches
Research, product, and engineering posts from the team building SolidMake.
What We Learned Shipping an AI-CAD Platform as a Final Year Project
Honest retrospective: what took longer than expected, what we'd sequence differently, and what surprised us about mixing ML research with product work.
— More posts
- 02Jul 16, 2026
Building a Studio: Designing the UX of AI-Assisted CAD
Before the model was ready, we built the interface it would eventually live behind — and shipped a working demo anyway.
productdesignfrontend - 03Jul 12, 2026
Inside the Round-Trip Filter: How We Guarantee Training Data Quality
About 10% of our transpiled data gets thrown away. That's not a bug — it's the whole point of the filter.
engineeringdataml - 04Jul 8, 2026
From Sketch to STEP: Designing the Image-to-CAD Pipeline
We shipped the cheaper image-to-CAD approach first, on purpose, and structured the API so upgrading it later costs nothing on the frontend.
engineeringaiarchitecture - 05Jul 4, 2026
Why We Chose CNC and Sheet-Metal Parts Over Organic Shapes
SolidMake will never generate a phone case or a figurine — that's a deliberate scope decision, not a limitation we're embarrassed about.
productcadstrategy - 06Jun 30, 2026
The Repair Loop: Teaching a Model to Fix Its Own Mistakes
When generated code fails to execute, we don't just fail the job — we feed the error back and let the model try again.
engineeringaibackend - 07Jun 25, 2026
Fine-Tuning a 1.5B Code Model on a Single Colab GPU
No cluster, no dedicated hardware — just QLoRA, a free-tier T4, and a lot of checkpointing discipline.
engineeringmltraining - 08Jun 20, 2026
Text2CAD and the Rise of Language-Model-Driven Engineering
Why a code-generation LLM, fine-tuned on real CAD sequences, is a better bet for parametric design than a diffusion model trained on renders.
researchaicad - 09Jun 15, 2026
What Makes a CAD Model "Manufacturable"? Our Validator Explained
A part that looks right in a 3D viewer can still be impossible to cut. Here's what we actually check before calling a generation a success.
engineeringmanufacturingcad - 10Jun 10, 2026
Executing Untrusted Model Code Safely: Our Sandbox Design
The model outputs Python. Running unreviewed, model-generated Python is a remote-code-execution risk unless you take it seriously.
engineeringsecuritybackend - 11Jun 5, 2026
From DeepCAD to CadQuery: Building the Transpiler
Turning 179,000 parametric construction sequences into training data meant building a transpiler first — and testing it before trusting it.
engineeringdatacad - 12Jun 1, 2026
Why CadQuery, Not Meshes: Choosing an Output Language for AI-Generated CAD
Most AI 3D tools generate meshes. We generate executable CadQuery Python — here's why that distinction is the whole point.
engineeringcadarchitecture
Want this in your inbox?
One email per release. No marketing fluff. Unsubscribe in one click.